net/bnxt: handle reset notify async event from FW
authorKalesh AP <kalesh-anakkur.purayil@broadcom.com>
Wed, 2 Oct 2019 01:23:23 +0000 (18:23 -0700)
committerFerruh Yigit <ferruh.yigit@intel.com>
Tue, 8 Oct 2019 10:14:30 +0000 (12:14 +0200)
When the FW upgrade is initiated the current instance
of FW issues a HWRM_ASYNC_EVENT_CMPL_EVENT_ID_RESET_NOTIFY
async notification to the driver. On receiving this notification,
the PMD shall quiesce itself and poll on the HWRM_VER_GET FW
command at regular intervals.

Once the VER_GET command succeeds, the driver should go through
the rediscovery process and re-initialize the device.

Also register with FW for the reset notify async event.

Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
drivers/net/bnxt/bnxt.h
drivers/net/bnxt/bnxt_cpr.c
drivers/net/bnxt/bnxt_cpr.h
drivers/net/bnxt/bnxt_ethdev.c
drivers/net/bnxt/bnxt_hwrm.c
drivers/net/bnxt/bnxt_hwrm.h
drivers/net/bnxt/hsi_struct_def_dpdk.h

index 37b4c71..8797b03 100644 (file)
@@ -333,6 +333,16 @@ struct bnxt_ctx_mem_info {
        struct bnxt_ctx_pg_info *tqm_mem[BNXT_MAX_TC_Q];
 };
 
+/* Maximum Firmware Reset bail out value in milliseconds */
+#define BNXT_MAX_FW_RESET_TIMEOUT      6000
+/* Minimum time required for the firmware readiness in milliseconds */
+#define BNXT_MIN_FW_READY_TIMEOUT      2000
+/* Frequency for the firmware readiness check in milliseconds */
+#define BNXT_FW_READY_WAIT_INTERVAL    100
+
+#define US_PER_MS                      1000
+#define NS_PER_US                      1000
+
 #define BNXT_HWRM_SHORT_REQ_LEN                sizeof(struct hwrm_short_input)
 struct bnxt {
        void                            *bar0;
@@ -463,6 +473,9 @@ struct bnxt {
        struct bnxt_ptp_cfg     *ptp_cfg;
        uint16_t                vf_resv_strategy;
        struct bnxt_ctx_mem_info        *ctx;
+
+       uint16_t                fw_reset_min_msecs;
+       uint16_t                fw_reset_max_msecs;
 };
 
 int bnxt_link_update_op(struct rte_eth_dev *eth_dev, int wait_to_complete);
index bbcdb42..1a23649 100644 (file)
@@ -4,6 +4,7 @@
  */
 
 #include <rte_malloc.h>
+#include <rte_alarm.h>
 
 #include "bnxt.h"
 #include "bnxt_cpr.h"
@@ -40,6 +41,21 @@ void bnxt_handle_async_event(struct bnxt *bp,
        case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_PORT_CONN_NOT_ALLOWED:
                PMD_DRV_LOG(INFO, "Port conn async event\n");
                break;
+       case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_RESET_NOTIFY:
+               /* timestamp_lo/hi values are in units of 100ms */
+               bp->fw_reset_max_msecs = async_cmp->timestamp_hi ?
+                       rte_le_to_cpu_16(async_cmp->timestamp_hi) * 100 :
+                       BNXT_MAX_FW_RESET_TIMEOUT;
+               bp->fw_reset_min_msecs = async_cmp->timestamp_lo ?
+                       async_cmp->timestamp_lo * 100 :
+                       BNXT_MIN_FW_READY_TIMEOUT;
+               PMD_DRV_LOG(INFO,
+                           "Firmware non-fatal reset event received\n");
+
+               bp->flags |= BNXT_FLAG_FW_RESET;
+               rte_eal_alarm_set(US_PER_MS, bnxt_dev_reset_and_resume,
+                                 (void *)bp);
+               break;
        default:
                PMD_DRV_LOG(INFO, "handle_async_event id = 0x%x\n", event_id);
                break;
index 8c6a34b..f48293b 100644 (file)
@@ -106,5 +106,6 @@ struct bnxt;
 void bnxt_handle_async_event(struct bnxt *bp, struct cmpl_base *cmp);
 void bnxt_handle_fwd_req(struct bnxt *bp, struct cmpl_base *cmp);
 int bnxt_event_hwrm_resp_handler(struct bnxt *bp, struct cmpl_base *cmp);
+void bnxt_dev_reset_and_resume(void *arg);
 
 #endif
index d90a6e4..6bc006a 100644 (file)
@@ -11,6 +11,7 @@
 #include <rte_ethdev_pci.h>
 #include <rte_malloc.h>
 #include <rte_cycles.h>
+#include <rte_alarm.h>
 
 #include "bnxt.h"
 #include "bnxt_cpr.h"
@@ -166,6 +167,8 @@ static int bnxt_vlan_offload_set_op(struct rte_eth_dev *dev, int mask);
 static void bnxt_print_link_info(struct rte_eth_dev *eth_dev);
 static int bnxt_mtu_set_op(struct rte_eth_dev *eth_dev, uint16_t new_mtu);
 static int bnxt_dev_uninit(struct rte_eth_dev *eth_dev);
+static int bnxt_init_resources(struct bnxt *bp, bool reconfig_dev);
+static int bnxt_uninit_resources(struct bnxt *bp, bool reconfig_dev);
 
 int is_bnxt_in_error(struct bnxt *bp)
 {
@@ -201,19 +204,25 @@ static uint16_t  bnxt_rss_hash_tbl_size(const struct bnxt *bp)
        return bnxt_rss_ctxts(bp) * BNXT_RSS_ENTRIES_PER_CTX_THOR;
 }
 
-static void bnxt_free_mem(struct bnxt *bp)
+static void bnxt_free_mem(struct bnxt *bp, bool reconfig)
 {
        bnxt_free_filter_mem(bp);
        bnxt_free_vnic_attributes(bp);
        bnxt_free_vnic_mem(bp);
 
-       bnxt_free_stats(bp);
-       bnxt_free_tx_rings(bp);
-       bnxt_free_rx_rings(bp);
+       /* tx/rx rings are configured as part of *_queue_setup callbacks.
+        * If the number of rings change across fw update,
+        * we don't have much choice except to warn the user.
+        */
+       if (!reconfig) {
+               bnxt_free_stats(bp);
+               bnxt_free_tx_rings(bp);
+               bnxt_free_rx_rings(bp);
+       }
        bnxt_free_async_cp_ring(bp);
 }
 
-static int bnxt_alloc_mem(struct bnxt *bp)
+static int bnxt_alloc_mem(struct bnxt *bp, bool reconfig)
 {
        int rc;
 
@@ -244,7 +253,7 @@ static int bnxt_alloc_mem(struct bnxt *bp)
        return 0;
 
 alloc_mem_err:
-       bnxt_free_mem(bp);
+       bnxt_free_mem(bp, reconfig);
        return rc;
 }
 
@@ -3523,6 +3532,89 @@ static const struct eth_dev_ops bnxt_dev_ops = {
        .timesync_read_tx_timestamp = bnxt_timesync_read_tx_timestamp,
 };
 
+static void bnxt_dev_cleanup(struct bnxt *bp)
+{
+       bnxt_set_hwrm_link_config(bp, false);
+       bp->link_info.link_up = 0;
+       if (bp->dev_stopped == 0)
+               bnxt_dev_stop_op(bp->eth_dev);
+
+       bnxt_uninit_resources(bp, true);
+}
+
+static int bnxt_restore_filters(struct bnxt *bp)
+{
+       struct rte_eth_dev *dev = bp->eth_dev;
+       int ret = 0;
+
+       if (dev->data->all_multicast)
+               ret = bnxt_allmulticast_enable_op(dev);
+       if (dev->data->promiscuous)
+               ret = bnxt_promiscuous_enable_op(dev);
+
+       /* TODO restore other filters as well */
+       return ret;
+}
+
+static void bnxt_dev_recover(void *arg)
+{
+       struct bnxt *bp = arg;
+       int timeout = bp->fw_reset_max_msecs;
+       int rc = 0;
+
+       do {
+               rc = bnxt_hwrm_ver_get(bp);
+               if (rc == 0)
+                       break;
+               rte_delay_ms(BNXT_FW_READY_WAIT_INTERVAL);
+               timeout -= BNXT_FW_READY_WAIT_INTERVAL;
+       } while (rc && timeout);
+
+       if (rc) {
+               PMD_DRV_LOG(ERR, "FW is not Ready after reset\n");
+               goto err;
+       }
+
+       rc = bnxt_init_resources(bp, true);
+       if (rc) {
+               PMD_DRV_LOG(ERR,
+                           "Failed to initialize resources after reset\n");
+               goto err;
+       }
+       /* clear reset flag as the device is initialized now */
+       bp->flags &= ~BNXT_FLAG_FW_RESET;
+
+       rc = bnxt_dev_start_op(bp->eth_dev);
+       if (rc) {
+               PMD_DRV_LOG(ERR, "Failed to start port after reset\n");
+               goto err;
+       }
+
+       rc = bnxt_restore_filters(bp);
+       if (rc)
+               goto err;
+
+       PMD_DRV_LOG(INFO, "Recovered from FW reset\n");
+       return;
+err:
+       bp->flags |= BNXT_FLAG_FATAL_ERROR;
+       bnxt_uninit_resources(bp, false);
+       PMD_DRV_LOG(ERR, "Failed to recover from FW reset\n");
+}
+
+void bnxt_dev_reset_and_resume(void *arg)
+{
+       struct bnxt *bp = arg;
+       int rc;
+
+       bnxt_dev_cleanup(bp);
+
+       rc = rte_eal_alarm_set(US_PER_MS * bp->fw_reset_min_msecs,
+                              bnxt_dev_recover, (void *)bp);
+       if (rc)
+               PMD_DRV_LOG(ERR, "Error setting recovery alarm");
+}
+
 static bool bnxt_vf_pciid(uint16_t id)
 {
        if (id == BROADCOM_DEV_ID_57304_VF ||
@@ -3965,6 +4057,22 @@ static int bnxt_setup_mac_addr(struct rte_eth_dev *eth_dev)
        return rc;
 }
 
+static int bnxt_restore_dflt_mac(struct bnxt *bp)
+{
+       int rc = 0;
+
+       /* MAC is already configured in FW */
+       if (!bnxt_check_zero_bytes(bp->dflt_mac_addr, RTE_ETHER_ADDR_LEN))
+               return 0;
+
+       /* Restore the old MAC configured */
+       rc = bnxt_hwrm_set_mac(bp);
+       if (rc)
+               PMD_DRV_LOG(ERR, "Failed to restore MAC address\n");
+
+       return rc;
+}
+
 static void bnxt_config_vf_req_fwd(struct bnxt *bp)
 {
        if (!BNXT_PF(bp))
@@ -4038,7 +4146,7 @@ static int bnxt_init_fw(struct bnxt *bp)
        return 0;
 }
 
-static int bnxt_init_resources(struct bnxt *bp)
+static int bnxt_init_resources(struct bnxt *bp, bool reconfig_dev)
 {
        int rc;
 
@@ -4046,9 +4154,15 @@ static int bnxt_init_resources(struct bnxt *bp)
        if (rc)
                return rc;
 
-       rc = bnxt_setup_mac_addr(bp->eth_dev);
-       if (rc)
-               return rc;
+       if (!reconfig_dev) {
+               rc = bnxt_setup_mac_addr(bp->eth_dev);
+               if (rc)
+                       return rc;
+       } else {
+               rc = bnxt_restore_dflt_mac(bp);
+               if (rc)
+                       return rc;
+       }
 
        bnxt_config_vf_req_fwd(bp);
 
@@ -4075,7 +4189,7 @@ static int bnxt_init_resources(struct bnxt *bp)
                }
        }
 
-       rc = bnxt_alloc_mem(bp);
+       rc = bnxt_alloc_mem(bp, reconfig_dev);
        if (rc)
                return rc;
 
@@ -4149,7 +4263,7 @@ bnxt_dev_init(struct rte_eth_dev *eth_dev)
                            "Failed to allocate hwrm resource rc: %x\n", rc);
                goto error_free;
        }
-       rc = bnxt_init_resources(bp);
+       rc = bnxt_init_resources(bp, false);
        if (rc)
                goto error_free;
 
@@ -4170,18 +4284,19 @@ error_free:
 }
 
 static int
-bnxt_uninit_resources(struct bnxt *bp)
+bnxt_uninit_resources(struct bnxt *bp, bool reconfig_dev)
 {
        int rc;
 
        bnxt_disable_int(bp);
        bnxt_free_int(bp);
-       bnxt_free_mem(bp);
+       bnxt_free_mem(bp, reconfig_dev);
        bnxt_hwrm_func_buf_unrgtr(bp);
        rc = bnxt_hwrm_func_driver_unregister(bp, 0);
        bp->flags &= ~BNXT_FLAG_REGISTERED;
        bnxt_free_ctx_mem(bp);
-       bnxt_free_hwrm_resources(bp);
+       if (!reconfig_dev)
+               bnxt_free_hwrm_resources(bp);
 
        return rc;
 }
@@ -4197,7 +4312,7 @@ bnxt_dev_uninit(struct rte_eth_dev *eth_dev)
 
        PMD_DRV_LOG(DEBUG, "Calling Device uninit\n");
 
-       rc = bnxt_uninit_resources(bp);
+       rc = bnxt_uninit_resources(bp, false);
 
        if (bp->grp_info != NULL) {
                rte_free(bp->grp_info);
index 24a5a09..abbad01 100644 (file)
@@ -27,6 +27,7 @@
 #include <rte_io.h>
 
 #define HWRM_CMD_TIMEOUT               6000000
+#define HWRM_SHORT_CMD_TIMEOUT         50000
 #define HWRM_SPEC_CODE_1_8_3           0x10803
 #define HWRM_VERSION_1_9_1             0x10901
 #define HWRM_VERSION_1_9_2             0x10903
@@ -97,6 +98,17 @@ static int bnxt_hwrm_send_message(struct bnxt *bp, void *msg,
                GRCPF_REG_KONG_CHANNEL_OFFSET : GRCPF_REG_CHIMP_CHANNEL_OFFSET;
        uint16_t mb_trigger_offset = use_kong_mb ?
                GRCPF_REG_KONG_COMM_TRIGGER : GRCPF_REG_CHIMP_COMM_TRIGGER;
+       uint32_t timeout;
+
+       /* Do not send HWRM commands to firmware in error state */
+       if (bp->flags & BNXT_FLAG_FATAL_ERROR)
+               return 0;
+
+       /* For VER_GET command, set timeout as 50ms */
+       if (rte_cpu_to_le_16(req->req_type) == HWRM_VER_GET)
+               timeout = HWRM_SHORT_CMD_TIMEOUT;
+       else
+               timeout = HWRM_CMD_TIMEOUT;
 
        if (bp->flags & BNXT_FLAG_SHORT_CMD ||
            msg_len > bp->max_req_len) {
@@ -139,7 +151,7 @@ static int bnxt_hwrm_send_message(struct bnxt *bp, void *msg,
        rte_write32(1, bar);
 
        /* Poll for the valid bit */
-       for (i = 0; i < HWRM_CMD_TIMEOUT; i++) {
+       for (i = 0; i < timeout; i++) {
                /* Sanity check on the resp->resp_len */
                rte_rmb();
                if (resp->resp_len && resp->resp_len <= bp->max_resp_len) {
@@ -151,7 +163,12 @@ static int bnxt_hwrm_send_message(struct bnxt *bp, void *msg,
                rte_delay_us(1);
        }
 
-       if (i >= HWRM_CMD_TIMEOUT) {
+       if (i >= timeout) {
+               /* Suppress VER_GET timeout messages during reset recovery */
+               if (bp->flags & BNXT_FLAG_FW_RESET &&
+                   rte_cpu_to_le_16(req->req_type) == HWRM_VER_GET)
+                       return -ETIMEDOUT;
+
                PMD_DRV_LOG(ERR, "Error(timeout) sending msg 0x%04x\n",
                            req->req_type);
                return -ETIMEDOUT;
@@ -657,12 +674,21 @@ int bnxt_hwrm_func_reset(struct bnxt *bp)
 int bnxt_hwrm_func_driver_register(struct bnxt *bp)
 {
        int rc;
+       uint32_t flags = 0;
        struct hwrm_func_drv_rgtr_input req = {.req_type = 0 };
        struct hwrm_func_drv_rgtr_output *resp = bp->hwrm_cmd_resp_addr;
 
        if (bp->flags & BNXT_FLAG_REGISTERED)
                return 0;
 
+       flags = HWRM_FUNC_DRV_RGTR_INPUT_FLAGS_HOT_RESET_SUPPORT;
+
+       /* PFs and trusted VFs should indicate the support of the
+        * Master capability on non Stingray platform
+        */
+       if ((BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)) && !BNXT_STINGRAY(bp))
+               flags |= HWRM_FUNC_DRV_RGTR_INPUT_FLAGS_MASTER_SUPPORT;
+
        HWRM_PREP(req, FUNC_DRV_RGTR, BNXT_USE_CHIMP_MB);
        req.enables = rte_cpu_to_le_32(HWRM_FUNC_DRV_RGTR_INPUT_ENABLES_VER |
                        HWRM_FUNC_DRV_RGTR_INPUT_ENABLES_ASYNC_EVENT_FWD);
@@ -683,14 +709,16 @@ int bnxt_hwrm_func_driver_register(struct bnxt *bp)
                 * this HWRM sniffer list in FW because DPDK PF driver does
                 * not support this.
                 */
-               req.flags =
-               rte_cpu_to_le_32(HWRM_FUNC_DRV_RGTR_INPUT_FLAGS_FWD_NONE_MODE);
+               flags |= HWRM_FUNC_DRV_RGTR_INPUT_FLAGS_FWD_NONE_MODE;
        }
 
+       req.flags = rte_cpu_to_le_32(flags);
+
        req.async_event_fwd[0] |=
                rte_cpu_to_le_32(ASYNC_CMPL_EVENT_ID_LINK_STATUS_CHANGE |
                                 ASYNC_CMPL_EVENT_ID_PORT_CONN_NOT_ALLOWED |
-                                ASYNC_CMPL_EVENT_ID_LINK_SPEED_CFG_CHANGE);
+                                ASYNC_CMPL_EVENT_ID_LINK_SPEED_CFG_CHANGE |
+                                ASYNC_CMPL_EVENT_ID_RESET_NOTIFY);
        req.async_event_fwd[1] |=
                rte_cpu_to_le_32(ASYNC_CMPL_EVENT_ID_PF_DRVR_UNLOAD |
                                 ASYNC_CMPL_EVENT_ID_VF_CFG_CHANGE);
@@ -837,7 +865,10 @@ int bnxt_hwrm_ver_get(struct bnxt *bp)
 
        rc = bnxt_hwrm_send_message(bp, &req, sizeof(req), BNXT_USE_CHIMP_MB);
 
-       HWRM_CHECK_RESULT();
+       if (bp->flags & BNXT_FLAG_FW_RESET)
+               HWRM_CHECK_RESULT_SILENT();
+       else
+               HWRM_CHECK_RESULT();
 
        PMD_DRV_LOG(INFO, "%d.%d.%d:%d.%d.%d\n",
                resp->hwrm_intf_maj_8b, resp->hwrm_intf_min_8b,
@@ -2685,6 +2716,10 @@ int bnxt_hwrm_func_qcfg(struct bnxt *bp, uint16_t *mtu)
        if (BNXT_VF(bp) && (flags & HWRM_FUNC_QCFG_OUTPUT_FLAGS_TRUSTED_VF)) {
                bp->flags |= BNXT_FLAG_TRUSTED_VF_EN;
                PMD_DRV_LOG(INFO, "Trusted VF cap enabled\n");
+       } else if (BNXT_VF(bp) &&
+                  !(flags & HWRM_FUNC_QCFG_OUTPUT_FLAGS_TRUSTED_VF)) {
+               bp->flags &= ~BNXT_FLAG_TRUSTED_VF_EN;
+               PMD_DRV_LOG(INFO, "Trusted VF cap disabled\n");
        }
 
        if (mtu)
index c882fc2..a036205 100644 (file)
@@ -21,6 +21,8 @@ struct bnxt_cp_ring_info;
        (1 << HWRM_ASYNC_EVENT_CMPL_EVENT_ID_PORT_CONN_NOT_ALLOWED)
 #define ASYNC_CMPL_EVENT_ID_LINK_SPEED_CFG_CHANGE      \
        (1 << HWRM_ASYNC_EVENT_CMPL_EVENT_ID_LINK_SPEED_CFG_CHANGE)
+#define ASYNC_CMPL_EVENT_ID_RESET_NOTIFY \
+       (1 << HWRM_ASYNC_EVENT_CMPL_EVENT_ID_RESET_NOTIFY)
 #define ASYNC_CMPL_EVENT_ID_PF_DRVR_UNLOAD     \
        (1 << (HWRM_ASYNC_EVENT_CMPL_EVENT_ID_PF_DRVR_UNLOAD - 32))
 #define ASYNC_CMPL_EVENT_ID_VF_CFG_CHANGE      \
index 0095717..809ea48 100644 (file)
@@ -8918,6 +8918,17 @@ struct hwrm_func_drv_rgtr_input {
         */
        #define HWRM_FUNC_DRV_RGTR_INPUT_FLAGS_ERROR_RECOVERY_SUPPORT \
                UINT32_C(0x20)
+       /*
+        * When this bit is 1, the function is indicating the support of the
+        * Master capability. The Firmware will use this capability to select
+        * the Master function. The master function will be used to initiate
+        * designated functionality like error recovery etc. If none of the
+        * registered PFs or trusted VFs indicate this support, then
+        * firmware will select the 1st registered PF as Master capable
+        * instance.
+        */
+       #define HWRM_FUNC_DRV_RGTR_INPUT_FLAGS_MASTER_SUPPORT \
+               UINT32_C(0x40)
        uint32_t        enables;
        /*
         * This bit must be '1' for the os_type field to be